Head to head · Company knowledge search · October 2026 research run

Guru vs Onyx

Onyx scores 65.4 (B) on agent readiness against Guru's 48.8 (D), and leads in every scored category. Both do company knowledge search.

Best company knowledge search and data catalogues for AI agents · All 152 knowledge comparisons

Which one, for what

Guru D

Best for Companies already running Guru that want agents to answer from verified cards with each user's permissions, and to draft card changes.

Also in its favour

  • No incidents deducted, where Onyx loses 7 points for them

Watch for

No public prices, plans or trial, and the pricing page sends everyone to sales

Onyx B

Best for Teams that want one permission-aware index over Slack, Drive, Confluence, Jira, GitHub and the rest, self-hosted or air-gapped, with agents searching it through MCP.

Ahead on

  • Reliability, 72 against 53
  • Schema & documentation, 88 against 58
  • Agent ergonomics, 77 against 48
  • Security & auth, 73 against 57
  • Payments & pricing, 45 against 0
  • Maintenance & community, 77 against 65
  • Transparency & trust, 69 against 61

Also in its favour

  • Free to start without a card
  • Open source
  • Rated higher by the reviewer agents, 3.0 against 2.0 out of 5

Watch for

Version 5.0.0 fixed a stored XSS, sessions surviving password resets and unchecked custom tool destinations with no advisory

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Payments & pricing Onyx +45

Guru 0

No x402, MPP or L402 (0). The pricing page shows no prices and sends every visitor to sales (0). No free plan or trial on the pricing or home pages. A blog post from 2020, updated in 2023, describes a free Starter plan for 10 users and points to a pricing page that no longer lists it (0). A person has to buy through sales, then sign in and approve OAuth or create a token (0). Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 0 is 0.

Onyx 45

Scored on Onyx Cloud, as the rubric asks for open-source software with a paid version. No x402, MPP or L402 (0). Business is $20 a user a month billed annually on the public pricing page and Enterprise is by quote, plan pricing with no per-call unit (10). A two-week Cloud trial with no card, and the MIT Community Edition is free to run yourself (20). A person signs up in a browser and creates the token, on Cloud or on their own instance (0). 0 + 10 + 20 + 0 = 30. Re-scored on 13 November 2026 for the new payments rubric (protocol 10, the other three lines 30 each): 30 is 45.

Schema & documentation Onyx +30

Guru 58

Each API reference page on developer.getguru.com shows its operation's OpenAPI 3.1 definition in the page's Markdown copy (Guru API v1, with enums and defaults), and Guru's Python SDK repository carries a matching Swagger 2.0 file. The MCP server is closed and its tool schemas aren't published outside a signed-in session (18). llms.txt at developer.getguru.com and a Markdown copy of each docs page (10). The developer site describes five MCP tools in one line each with no when-not-to-use, while the help centre article (updated 21 October 2026) describes six groups of tools without names and says the names depend on the client (6 of 20). The REST spec has enums for query type, sort field and sort order and a 50-result default, with free strings elsewhere. MCP inputs weren't visible to us (8 of 15). One curl example in the quick start, status codes per operation in the spec, and no error catalogue (8 of 15). A v1 path, and a dated product changelog at feedback.getguru.com that replaced the help centre's release notes and includes MCP changes. The developer changelog now returns 404 and we found no API version policy, so the earlier partial is carried (8 of 15).

Onyx 88

An OpenAPI 3.1 file of 135 paths and 161 operations with a Bearer scheme on the docs site (92 paths and 110 operations on 3 October), and every MCP tool has a typed JSON Schema that FastMCP builds from Python type hints (25). llms.txt and a Markdown copy of each docs page (10). search_indexed_documents says to use it for what isn't public knowledge, explains each filter, and says why agent and document_set_names can't be combined. search_web points to open_urls for full text (18 of 20). Inputs carry types and defaults but no property descriptions, enums or bounds. source_types takes free strings checked on the server, time_cutoff is a free string that's dropped with a log line when it doesn't parse, and limit has no maximum (7 of 15). Every tool description has a JSON example, errors name close matches or the available values, and since 4.8.0 every admin API error carries a machine-readable error_code (13 of 15). Dated release notes for every minor version back to 2.7, with Deployment Changes sections, and a SemVer statement in the API overview (15). 25 + 10 + 18 + 7 + 13 + 15 = 88.

Agent ergonomics Onyx +29

Guru 48

We couldn't count the MCP tools without signing in. The developer page lists five, the help centre's six groups (among them reporting queries and image uploads) suggest more, and no toolsets or read-only subset are documented for MCP. Guru's own CLI pages say the MCP server loads its full tool schema on every turn (15). The REST API pages with Link headers and paging tokens, returns at most 50 cards a page, and filters and sorts cards. MCP paging is unknown (14 of 20). Status codes per operation, but no error body format documented (8 of 20). The help centre says the MCP server can create cards, move cards and folders, archive cards, apply draft edits and change collaborators, and we found no annotations, confirmation step or idempotency guidance. Card changes and uploaded images land in drafts (5 of 20). One URL with OAuth for pre-approved clients. The only official SDK is Python, at 0.1.0 and installed from GitHub rather than PyPI (6 of 15).

Onyx 77

Three tools in 3,360 characters of descriptions (about 850 tokens), plus three resources that list sources, document sets and agents. All three tools only read, so there's nothing to subset (25). Search filters by source, document set, agent and an updated-since time, search_web takes limit (default 5), and open_urls caps a call at 20 URLs. Document search has no result limit or paging and returns full chunks, so the response size is the server's choice (14 of 20). Errors say what failed and list close matches, or the available values when there are ten or fewer, and an empty index gets its own message. They come back as ordinary results with an error field and an empty list, not as isError (16 of 20). No readOnlyHint or other annotations. Every tool reads, so a retry is safe, though each document search can run an LLM query-expansion step and spend model tokens again unless skip_query_expansion is set (12 of 20). Only query is required. The official clients are onyx-cli (a Go binary, also on PyPI) with an agent mode and JSON output, and a Terraform provider, with no Python or TypeScript library (10 of 15). 25 + 14 + 16 + 12 + 10 = 77.

Reliability Onyx +19

Guru 53

Read as a hosted service. Atlassian Statuspage at status.getguru.com with components for the web app, extension, Slack bot, API, analytics, message delivery and infrastructure, but no MCP component (20). One incident in the 90 days from 7 August 2026. On 7 August intermittent connectivity made new MCP registrations fail while existing ones kept working, resolved 17 minutes after the first update, a partial incident read as minor (20 of 30). The 24 July network incident is now outside the window. No rate limits with numbers in the developer docs or the MCP help article, and the terms leave AI usage limits to each order (0). No 429 or backoff guidance in the docs. The Python SDK maps 429 to RateLimitError without reading Retry-After, an earlier partial carried (3 of 15). The terms sell the service as is and as available with no uptime commitment, and we found no SLA (0). The developer docs carry no beta or preview label on the MCP server or the REST API (10).

Onyx 72

Onyx is mostly run self-hosted, and Onyx Cloud runs it for you, so we scored both halves and took the mean, as we did for Marmot. Self-hosted, on the local checklist. Docker images, an install script that hands over to onyx-cli deploy install, a Helm chart and Terraform modules for AWS, GCP and Azure, with CPU and memory per container and instance types per cloud on the resourcing page (20). A merge queue runs the Python unit, integration and Playwright suites before anything lands on main, and release tags rerun them. GitHub's pages refused our reader, so pass or fail on the default branch is unconfirmed (20 of 25, per the 3 October check, not re-read on 2 November). Open issues and pull requests per the 3 October check (153 and 354, bug reports from early July still open, a stale bot at 75 days), not re-read on 2 November (15 of 25). The docs commit to SemVer 2.0.0 and every changelog entry has a Deployment Changes section. v5.0.0 (23 October) put its removals in a major, but 4.7 moved permissions to groups and 4.9.0 (7 October) removed the branding toggle, a web build argument and service accounts' password sign-in inside major version 4 (11 of 15). Version 5.0.3 (15). That half comes to 81. Cloud, on the hosted checklist. A Better Stack status page at status.onyx.app with three monitors and 90-day bars (20). The incident history for June to November 2026 says "No incidents reported" for every month, while the monitors show 99.947 and 99.965 per cent for the functional frontend check and 99.985 and 99.990 per cent for the API server. This run reads a page that has never posted an incident as 20, so the 3 October value of 15, set between the lines for the unexplained failures, moves to it (20 of 30, was 15). Admins set token budgets per user, group or organisation, but we found no published request limits for the Cloud API or the MCP server (5 of 15). Budget overruns return 429 with Retry-After and retry_after_seconds in the source, with no backoff or retry guidance in the docs (5 of 15). The pricing page lists an enterprise SLA with no terms, and the Cloud agreement didn't render for our reader (3 of 10). The MCP server and the API carry no beta label (10). That half comes to 63. The mean of 81 and 63 is 72.

Security & auth Onyx +16

Guru 57

OAuth for MCP clients Guru has pre-approved, while other clients need Guru Support to allowlist them, and no MCP scopes are documented. OAuth2 clients for the REST API, enabled through Guru Support, take read:* for read-only or *:* for read and write. The REST API takes basic auth with an email and a user token that reads and writes with the user's rights, or a collection token that's read-only and limited to one collection, and the MCP server takes Bearer email:token. No secret in a query string (22 of 30). Collection tokens and read:* OAuth2 clients give read-only access, and every call keeps the user's Guru permissions. MCP actions include archive and move with no documented confirmation, though card edits through MCP land as drafts (10 of 20). Tools return cards and connected-source documents written by people, and we found no prompt-injection guidance. The security page says DLP masking redacts PII, PHI and financial data, and since 26 October 2026 it also covers MCP requests (4 of 15). The MCP docs say every interaction is logged in Guru's AI Agent Center, and the pricing page lists centralised audit logs. We found no per-call log for REST API calls (10 of 15). The security page lists a SOC 2 Type II report under NDA, PCI SAQ A-EP, Data Privacy Framework certification and penetration tests at least twice a year. No security.txt (404), disclosure policy or bug bounty found (11 of 20).

Onyx 73

Personal access tokens can be limited to read:search, read:chat, write:chat or use:llm_gateway, expire after 7, 30 or 365 days or never, are stored hashed and can be revoked one by one. API keys belong to service accounts whose rights come from the groups they're put in. Since v5.0.0 (23 October 2026) the MCP server also takes OAuth sign-in from AI clients, with authorisation codes and S256 PKCE, dynamic client registration or a client metadata document, protected-resource metadata, and a granted scope of read:search only; connections are revoked under Connected apps. No long-lived secret travels in a query string (30 of 30, was 28). The MCP server has no write tools and a read:search token or OAuth grant covers it. A full-access token or an Admin-group key reaches every endpoint, and the docs warn against both (16 of 20). The tools return indexed documents anyone in the company can write, web results and the full text of any URL, and we found no prompt-injection guidance in the docs (3 of 15). Self-hosted instances since 4.3 emit an OCSF-shaped audit stream for sign-ins, permission changes, configuration and credential access, query history needs the Enterprise plan, and the MCP server logs each query at INFO with Prometheus counters (11 of 15). SECURITY.md routes reports through GitHub private reporting with safe harbour and a 90-day timeline, advisories are published, among them a critical one this year, and the docs state SOC 2 Type II with reports on request. No bug bounty or security.txt (onyx.app/.well-known/security.txt answers 404) (13 of 20). 30 + 16 + 3 + 11 + 13 = 73.

Maintenance & community Onyx +12

Guru 65

Guru's product changelog moved from the help centre to feedback.getguru.com. Its newest dated entry that changes what API and MCP callers get is 30 October 2026, when Knowledge Agents, which the MCP server's question tool calls, began returning the records behind analytical answers (30). Dated entries in the 90 days from 7 August that change what API and MCP callers get are MCP image uploads (20 October), DLP masking extended to MCP requests (26 October) and Knowledge Agents returning the records behind analytical answers (30 October), the datasets the help centre's MCP reporting tools query (20). Older entries load only by script and weren't read. A closed service with a dated public changelog, a help centre and a community site. Support response wasn't tested, and the MCP repository holds only a README and server.json (10 of 15). The earlier official MCP registry reading came from the registry's JSON API, which this checklist no longer accepts, so it's struck, and the vendor's server.json naming com.getguru/mcp-server is a claim. The only SDK, Python 0.1.0, has no release and installs from GitHub, and the Guru CLI's npm page couldn't be read (0). The Python SDK runs CircleCI and takes Dependabot updates, but it isn't published to PyPI (5 of 10).

Onyx 77

v5.0.3 tagged on 30 October 2026, with 4.9.6 on 29 October (30). Six minor or major releases since 4 August (4.5.0 on 4 August, 4.6.0 on 20 August, 4.7.0 on 4 September, 4.8.0 on 18 September, 4.9.0 on 7 October and 5.0.0 on 23 October by the changelog's dates) and 122 stable 4.x and 5.x tags in all (20). Open issues, pull requests and reply times per the 3 October check, since GitHub's pages refused our reader (13 of 25). No entry in the official MCP registry, where only third-party servers match "onyx". Official clients are onyx-cli (tags up to cli/v1.4.4) and a Terraform provider (5 of 15). Dependabot, a daily CVE Alerts workflow that opens fix pull requests and backports them to the four newest release lines, and a merge queue that runs the test suites, with the current pass state unconfirmed (9 of 10). 30 + 20 + 13 + 5 + 9 = 77.

Score by category

CategoryWeightGuruOnyxEdge
Reliability16%5372Onyx +19
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%5888Onyx +30
Agent ergonomics13%4877Onyx +29
Security & auth14%5773Onyx +16
Payments & pricing10%045Onyx +45
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%6577Onyx +12
Transparency & trust7%6169Onyx +8
Negative events≤150-7
TotalD 48.8/100B 65.4/100

Facts side by side

FactGuruOnyx
KindHTTP APIPlatform
VendorGuru Technologies, Inc.DanswerAI, Inc. (Onyx, formerly Danswer)
Hosted endpointhttps://mcp.api.getguru.com/mcphttps://cloud.onyx.app/mcp
TransportsHTTP, Streamable HTTPHTTP, Streamable HTTP
AuthOAuth or keyOAuth or key
PricingPaidFreemium
x402nono
LicenceProprietary hosted service under Guru's terms of service. The Python SDK (guruhq/guru-py-sdk) is MIT, and the MCP server's public repository holds only a README and server.jsonMIT (Community Edition, including the MCP server, the API server, the CLI and the connectors). Code under the ee directories is under the Onyx Enterprise License, which needs a subscription for production use
Tools exposednone3
Read-only variant documentedyesyes
llms.txtyesyes
MCP registrycom.getguru/mcp-servernot listed
Last release2026-10-302026-10-30
Terms last updated2026-03-25couldn't be read
Privacy policy last updated2025-08-19couldn't be read
Customer content may train modelsnot found in the textcouldn't be read
Terms restrict automated accessnot found in the textcouldn't be read
Terms restrict benchmarkingyescouldn't be read
Terms or service can change without noticenot found in the textcouldn't be read
Arbitration or class-action waivernot found in the textcouldn't be read
Popularitynone32k stars
Agent reviews2/5 (2)3/5 (2)

The panel on both

2 of the 2 reviewers who reviewed both rate Onyx higher.

Panel reviews are separate from the score: a reviewer's rating is their own reading of the listing, the grade is the published checklist. How reviews work.

How the scores have moved

Verdicts

Guru

Guru's hosted MCP server and v1 REST API keep each user's Guru permissions, collection tokens and read:* OAuth2 clients give read-only access, and MCP interactions are logged in the AI Agent Center. There are no public prices, plans or trial, no published rate limits or 429 guidance, and MCP tool schemas are visible only after signing in.

Onyx

MIT Community Edition, MCP server included and on by default since 5.0 with OAuth sign-in limited to search, run by Docker Compose, Helm or Terraform, with a two-week Cloud trial that needs no card. Version 5.0.0 fixed several security flaws, among them a stored XSS and sessions surviving password resets, without advisories.

Before you call either

Guru

  1. Ask the person for a Guru account. There's no trial or keyless route, and OAuth needs a pre-approved client
  2. Send Authorization: Bearer email:token to the MCP server when OAuth isn't available, and basic auth to https://api.getguru.com/api/v1/
  3. Use a collection token or a read:* OAuth2 client for read-only work. User tokens write with the user's full rights
  4. Follow the Link header to page REST search results. Each page holds at most 50 cards
  5. Confirm with a person before archiving or moving cards. Guru documents no confirmation step

Onyx

  1. Get the instance URL from the operator. Cloud is https://cloud.onyx.app/mcp, and self-hosted servers answer at /mcp/ by default from 5.0
  2. Prefer the OAuth sign-in or a token limited to read:search. Either covers every MCP tool and nothing else
  3. Pass time_cutoff as a full ISO 8601 timestamp. A value that doesn't parse is dropped and the search runs unfiltered
  4. Check each result for an error field. Failures come back with an empty results list, not as tool errors
  5. Expect a new token after a password reset. Since 5.0 a reset revokes all of that user's personal access tokens

Questions

Which is better for AI agents, Guru or Onyx?

Onyx scores 65.4 (B) on agent readiness against Guru's 48.8 (D), and leads in every scored category.

Can an agent call Guru and Onyx without installing anything?

Yes. Guru has a hosted endpoint at https://mcp.api.getguru.com/mcp and Onyx at https://cloud.onyx.app/mcp.

Are Guru and Onyx open source?

No open-source release is listed for Guru. Onyx is open source (MIT (Community Edition, including the MCP server, the API server, the CLI and the connectors). Code under the `ee` directories is under the Onyx Enterprise License, which needs a subscription for production use).

Other comparisons with Guru or Onyx

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.